How can I calculate the metrics for a GAN model: Accuracy, precision, F1, ROC and confusion matrix?
Hi Najeh…
Its rather unfortunate no one has replied after all this while. This is for anyone else in the future, just in case.
GANs don’t converge (meaning there’s no clear endpoint to training where one will say it can no longer improve). And the outputs are not of the type that can be directly quantitatively assessed using F1, etc.
GANs can be assessed by other means such as the Frechet Inception Index (FID), Inception Score, etc for measuring generated image quality.
You can refer to the following github for help in evaluating your GAN: w86763777/pytorch-image-generation-metrics: Pytorch implementation of common image generation metrics. (github.com)